Students performance clustering for future personalized in learning virtual reality

Abstract

This study investigates five clustering algorithms—K-Means, Gaussian mixture model (GMM), hierarchical clustering (HC), k-medoids, and spectral clustering—applied to student performance in mathematics, reading, and writing to support the development of virtual reality (VR)-based adaptive learning systems. Cluster quality was assessed using Davies-Bouldin and Calinski-Harabasz indices. Spectral clustering achieved the best results (DBI = 0.75, CHI = 1322), followed by K-Means (DBI = 0.79, CHI = 1398), while HC demonstrated superior robustness to outliers. Three distinct student profiles—beginner, intermediate, and advanced—emerged, enabling targeted adaptive interventions. Supervised classifiers trained on these clusters reached up to 99% accuracy (logistic regression) and 97.5% (support vector machine (SVM)), validating the discovered groupings. This work introduces a novel, data-driven methodology integrating unsupervised clustering with supervised prediction, providing a practical framework for designing immersive VR learning environments.